Messages in this thread |  | | Date | Tue, 16 Oct 2001 21:09:51 -0400 (EDT) | Subject | RAID 5 performance better in degraded mode!? | From | "Adam Kramer" <> |
| |
Hi all,
If this has been been covered already I apologize in advance, I spent several hours looking over the archives and could find nothing on this particular issue.
(I am running kernel 2.4.12)
I have 4 WD600AB drives (I'll call them hda hdc hde hdg) which I want to run in RAID 5. I have one drive per channel, 2 connected to the motherboard (VIA KT266A chipset) and 2 connected to a Promise Ultra66 controller. Each drive is capable of ~38MB/sec sequential transfer rate. My primitive method for benchmarking is to run vmstat 1 on a virtual console and cat /dev/hda > /dev/null on another, and look at how many blocks it reads per second. Sloppy yes, but the results are just about exactly the same as bonnie reports and this takes less time :)
The promise driver seems to have some performance issues, if I cat /dev/hde > /dev/null & cat /dev/hdg > /dev/null & and look at blocks in, it maxes out around 53MB/sec, if I do the same with hda and hdc I get ~77MB/sec which follows with how fast the drives can transfer. If I read from all 4 hard drives at the same time i'm pulling about 125MB/sec total from the drives, which seems to indicate to me it isn't a hardware/IDE driver problem.
I made an array in degraded mode on hdc hde and hdg intending to copy files over and then add hda into the array as well. I tested reads from the md device and was getting ~53MB/sec from it, which seemed about right since 2 of the drives were connected to the promise controller.
I then copied my data over from hda onto the array, and added it. Once the array was done reconstructing, the fastest I could read off of it was ~44MB/sec maximum, average around 41. Writing was actually faster, on the degraded array writes were about 28MB/sec and on the 4 drive array they are 40MB/sec. I expected this to be the opposite, writing might be faster on a degraded array but reading should definitely be slower.
I am using a stripe size of 64K, but tried it with a 128K stripe and it made no difference.
I made a 3 drive raid0 array out of hdc hde hdg and was reading at 76MB/sec off of it which seems much more reasonable.
I tried 2.4.12-ac2 and had SERIOUS performance issues with the hard drives, things just didn't "feel" right.
I am not yet subscribed to linux-kernel so please CC me if anyone has any interesting ideas.
thanks, Adam Kramer
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
|  |